been cleaning up after lambda migrations for two years now. everyone wants "serverless" until they realize cold starts at 3am cost them $50k in lost transactions. or they hit the 15-minute timeout on a job that legitimately needs 45 minutes.
the real issue: lambda encourages lazy architecture. you throw business logic into handlers, skip proper error handling because "it's managed", then wonder why your dlq is full of unparseable messages. saw a team store entire 200mb datasets in /tmp because they didn't want to design a proper cache layer.
// what everyone writes
handler = (event) => {
db.query(big_expensive_query)
process_for_20_minutes() // timeout incoming
return result
}
if you're doing synchronous work over 5 minutes or dealing with stateful connections, just use fargate or ec2. the cost difference doesn't matter if your system actually works. lambda shines for events and webhooks, not general purpose compute.
the 50ms p99 latency everyone quotes. sure. if you're not doing anything real.
No responses yet.